Multiscale Meets Spatial Awareness: An Efficient Attention Guidance Network for Human Parsing

Table 3

Human parsing results with fifteen state-of-the-art methods on the LIP validation set.

Method	mIoU
SegNet [44]	18.17
FCN-8s [19]	28.29
DeepLab (VGG16) [10]	41.64
Attention [17]	42.92
DRN-50 + Vortex [45]	41.09
DeepLab (ResNet-101) [10]	44.80
SS-JPPNet [3]	44.73
SS-NAN [16]	47.92
SPReID [9]	48.16
MuLA [6]	49.30
CE2P [25]	53.10
BraidNet [23]	54.40
HRNetV2 [46]	56.48
CNIF [24]	57.74
A-CE2P [47]	59.36
Baseline (VGG16)	40.41
Baseline + Attention ASPP	42.31
Baseline + Attention RefineNet	43.78
AG-Net (VGG16)	46.33
AG-Net (DenseNet-121)	50.54

The Baseline is AG-Net without the attention mechanism proposed in this paper.